ArtiÞcial neural network
Since the gain-in nodeÕs input-output transformation
function determines the slope of the sigmoid curve, it
a¤ects the networkÕs prediction accuracy for the test set
(shown in Table 5). It was found that the network using
the leave-one-out procedure gave satisfactory performance
by using a gain value between 0.5 and 1.1 when
the number of hidden nodes is 20. In order to obtain an
optimum ANN model, various network architectures
were tested as shown in Table 6. The number of hidden
nodes (N) was an adjustable parameter that was
optimized by reducing the number used until the network
Õs prediction performance deteriorated or the best
prediction accuracy was found. It was found that 10 to
30 hidden nodes plus 1 bias node gave the best prediction
performance for the wine samples.
With a gain of 0.8 and 10 to 30 hidden nodes, the
ANNachieves good performance with a prediction rate
of 100%. It must be pointed out that the nodeÕs inputoutput
transformation function has a considerable effect
on the performance of ANN. If one adopts the
input-output transformation functions: f(x)
"1/(1#e(~9@h)) for the output layer and f(x)
"2/(1#e(~29@h))!1 for the hidden layer, the best
prediction rate of the ANN is only about 94.7% with
a gain of 1.0 and 20 hidden nodes (Table 7). Table 4